# Query Fan-Out Explained: Why One Prompt Isn't Enough

> Query fan-out is how Google's AI Mode and AI Overviews build an answer. They split a question into subtopics, run many related searches at once and write one response from the pages those searches return. Ranking for the main prompt isn't enough. You need a clear page or source for each sub-question buyers trigger.

By Othmane Khadri, founder of Earleads. Published 2026-10-21. Canonical: https://www.earleads.com/blog/query-fan-out/

## What is query fan-out?

Query fan-out is the step between a person's question and the AI answer. Instead of running one search for the words typed, Google's AI search splits the question into subtopics, runs many searches at once and writes one response from the pages those searches return.

You don't need to guess at this. Google has described it in four places, and they line up.

| Date | Google source | What it says |
|---|---|---|
| March 5, 2025 | [AI Mode launch post](https://blog.google/products/search/ai-mode-search/) | AI Mode "uses a 'query fan-out' technique, issuing multiple related searches concurrently across subtopics and multiple data sources and then brings those results together" |
| May 20, 2025 | [I/O 2025 update](https://blog.google/products/search/google-search-ai-mode-update/) | AI Mode is "breaking down your question into subtopics and issuing a multitude of queries simultaneously on your behalf." Deep Search uses the same technique and "can issue hundreds of searches" |
| July 2025 | Robby Stein, Google's VP of Product for Search, [quoted by Search Engine Journal](https://www.searchenginejournal.com/query-fan-out-technique-in-ai-mode-new-details-from-google/552532/) | For "things to do in Nashville with a group," it "may think of a bunch of questions like great restaurants, great bars, things to do if you have kids, and it'll start Googling basically" |
| December 10, 2025 | [Search Central, AI features and your website](https://developers.google.com/search/docs/appearance/ai-features) | Both AI Overviews and AI Mode "may use a 'query fan-out' technique," issuing multiple related searches across subtopics and data sources to develop a response |

Stein's example is the clearest one. Nobody typed "great bars in Nashville." The engine decided that a good answer about a group trip needed it, searched for it and picked sources for it. The person sees one answer. Behind it sit several searches, each with its own set of winning pages.

The same thing happens on buying questions. A finance lead asking which payroll software suits a 40-person company may never type "payroll software pricing" or "does it integrate with our HR tool." The engine may search both anyway, because a complete answer needs them.

Other AI assistants search the web too, and Google's Gemini API describes the same basic move for developers, where the model "automatically generates one or multiple search queries and executes them" ([Google AI for Developers](https://ai.google.dev/gemini-api/docs/google-search)). Google has described fan-out for its own search product in the most detail, so this guide uses its words.

## Why winning one prompt isn't enough

In classic search, you could pick a keyword, build the best page for it and measure your rank. Fan-out breaks that model in two ways.

First, the answer pulls from more pages than one results page would show. Google says that while responses are generated, its models "identify more supporting web pages, allowing us to display a wider and more diverse set of helpful links associated with the response than with a classic web search" ([Google Search Central](https://developers.google.com/search/docs/appearance/ai-features)). A wider set of links means more seats at the table, and they go to whoever wins the sub-searches.

Second, the pages that win the sub-searches are the ones that tend to get cited. Surfer pulled the top 10 pages for 10,000 keywords, used Gemini to generate 33,000 fan-out queries for the ones with AI Overviews and scraped the top pages for those too. Of the AI Overview citations that ranked in organic search, 51.2% ranked for the main query and at least one fan-out query. Another 29.2% ranked only for fan-outs, and just 19.6% ranked for the main query alone. The correlation between the number of fan-outs a page ranked for and its chance of citation was 0.77 ([Surfer, September 2026](https://surferseo.com/blog/query-fan-out-impact)).

Two caveats keep that honest. It's a correlation, and Surfer's fan-outs came from Gemini, which approximates what Google runs rather than recording it. The same study also found 67.82% of AI Overview citations didn't rank in the top 10 for the main query or any fan-out, so rankings explain only part of who gets cited. Ahrefs saw the same drift from the other side, with far fewer cited pages ranking in the top 10 in March 2026 than in mid-2025. We cover that study in [AI search ranking factors](/blog/ai-search-ranking-factors/).

The practical reading is simple. If you win the head prompt and nothing around it, you've won one of the searches behind the answer. A competitor with weaker head-term rankings but solid pages on pricing, integrations and switching can still be the brand the answer leans on.

## What query fan-out doesn't mean

A new mechanism attracts a lot of bad advice. Three claims are worth clearing up before any work starts.

**It doesn't need special markup.** Google says "there are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary." It also says "you don't need to create new machine readable files, AI text files, or markup to appear in these features," and there's no special schema to add. A supporting link must be "indexed and eligible to be shown in Google Search with a snippet" ([Google Search Central](https://developers.google.com/search/docs/appearance/ai-features)). Fan-out changes which pages get fetched, not the entry ticket.

**It doesn't mean a page per sub-query.** Fan-out searches overlap. Four pages on price, cost, "how much is it" and a pricing guide answer one question four times, and each one ends up weaker than a single good page. The skill is deciding which sub-questions deserve their own URL.

**It isn't a list you can download.** Fan-out depends on the question, the person and the moment. Every "query fan-out tool" on the market simulates it, mostly by asking a Gemini model with search grounding what it would search. That's useful for ideas. It isn't a record of what Google ran for your buyer yesterday.

## Step 1: Start from buyer questions, not keywords

Fan-out begins with the question a person asks, so your map should too. A keyword tool shows what people typed into a search box. It can't show the longer, layered questions people now ask an assistant, and those are the ones that fan out the most.

At [Earleads](/), every client's tracked set is built from positioning and sales language, not from a keyword tool. Each question is run live, and it stays in the set only if an answer exists and the client is missing from its sources. If no answer exists, there's nothing to win. If the client is already cited everywhere, there's nothing to fix.

Good sources for those questions:

- **Sales call notes and demo recordings.** The questions buyers ask before they buy are the ones they ask an assistant first.
- **Support tickets and onboarding emails.** These carry the "will it work for me" questions that sit behind every shortlist.
- **Lost deal reasons.** If prospects keep choosing a rival for the same reason, that comparison is a question buyers put to AI.
- **What customers say at signup.** A "how did you hear about us" answer often repeats the exact question that led them to you.

Write each one the way a buyer would say it. "Best payroll software for a 40-person company with staff in two countries" fans out very differently from "payroll software." For a consumer brand, "best running shoes for flat feet if I run on trails" is a different question from "running shoes," with different sub-searches behind it.

## Step 2: Expand each question into its sub-questions

Now turn each buyer question into a family. You're asking what a careful researcher would need to look up to answer it well, because that's roughly what the engine does.

Use three inputs and compare them:

1. **A fan-out simulation.** Run the question through a query fan-out tool, or through Gemini with search grounding, which returns the searches it used. Run it several times. Sub-questions that appear in most runs are strong candidates. One-offs are noise until you see them again.
2. **What Google already shows.** The "People also ask" questions, related searches and the sources in today's AI Overview for that question all hint at the subtopics Google associates with it.
3. **What your buyers actually need to decide.** Most families share the same facets, and you can fill them in from sales knowledge.

| Facet | B2B example (payroll software) | B2C example (trail running shoes) |
|---|---|---|
| Price | Cost for 40 employees, per-country fees | Price range, best value pick |
| Fit | Works for companies with staff abroad | Good for flat feet, wide feet |
| Comparison | You against a named rival | Two named models side by side |
| Proof | Ratings, reviews, customer stories | Reviews from runners with the same problem |
| Setup and switching | Migration from a current tool, setup time | Break-in time, sizing versus another brand |
| Risk | Compliance, data residency | Return policy, durability |

Keep the family short enough that each sub-question can have an owner. A family of forty isn't a plan, it's a spreadsheet nobody reads.

## Step 3: Assign every sub-question to a surface

This is the step most fan-out guides skip. Knowing the sub-questions doesn't tell you where the answer should live, and not every answer belongs on your own site.

![A buyer question about payroll software for a 40-person company fans out into six sub-searches. Price, comparison with a rival and integrations point to the brand's own site. Ratings and best picks for small teams point to third-party review pages and lists. Whether setup was painful points to forum threads. A yellow note says every sub-question gets one owner, one surface and one page to point at](/blog/query-fan-out/diagram-1.svg)
*A buyer question fans out into sub-searches, and each sub-search is won on a different kind of page.*

At Earleads, each tracked question is assigned to a surface before anything is written. First-person experience questions ("was setup painful," "did it hold up after six months") go to community threads, because an engine looking for lived experience looks there. Named comparisons go to the client's own site, where the brand controls the facts. The same logic works for any team:

- **Your site owns facts you control.** Pricing, integrations, limits, comparisons where you're one of the two names, migration guides.
- **Third-party pages own judgement.** Review sites, "best of" lists and industry write-ups answer "which is best" and "is it any good." You earn a place there with accurate listings and concrete detail, not by writing the list yourself.
- **Forums and online communities own experience.** "Has anyone switched from X" and "is it worth it" are answered by people. A useful reply from a named person with real history in that community is the version that holds up.

Write the surface next to each sub-question. If two sub-questions point at the same page, that's fine. If one sub-question points at nothing, you've found the gap that loses you the answer.

## Step 4: Build site architecture that matches the family

Once you know which sub-questions your site owns, shape the site around them. The structure that works for fan-out is hub and spokes.

![On the left, four grey pages target payroll software price, payroll software cost, how much is payroll software and a payroll pricing guide, and together they split one answer. On the right, a yellow hub page for payroll software for small companies links to spoke pages for pricing, a comparison with a named rival, integrations and switching from a rival, so each sub-search lands on one clear page](/blog/query-fan-out/diagram-2.svg)
*Variants of one question belong on one page. Distinct sub-questions get their own spoke, linked to the hub.*

The hub owns the head question, the way a page on "payroll software for small companies" or "trail running shoes for flat feet" would. The spokes own sub-questions with a different answer and a buyer who could act on it. That's the same rule we use for comparison pages. The hub owns the head term ("X alternatives"), and the spokes are qualifiers like "for agencies," "cheaper" or "migrating from."

A sub-question gets its own spoke when three things are true. The answer is different from the hub's. The page could carry a table or facts no other page has. And a buyer reading it could take a next step. Everything else becomes a section on the hub with the sub-question as its heading.

Then link the family together. Google's own best practices for AI features include "making your content easily findable through internal links on your website" and "making sure that important content is available in textual form" ([Google Search Central](https://developers.google.com/search/docs/appearance/ai-features)). Link hub to spokes and spokes back to the hub, with anchor text that names the sub-question. Keep prices, limits and specs in HTML text, not inside images, PDFs or tabs that load later. If crawlers can't read the page at all, none of this matters, so check that first with our guide to [blocking AI crawlers by accident](/blog/blocking-ai-crawlers-by-accident/).

The same rule of one topic, one place holds off your site. Two threads on the same question compete with each other, and the one that wins may not be the one with your best answer in it.

## Step 5: Write passages each sub-search can lift

A fan-out search for "payroll software integrations with HR tools" doesn't need your whole page. It needs the paragraph that answers that question. So write the page as a set of answers.

- **Use the sub-question as the heading.** "Does it work with our HR tool?" is easier to match than "Ecosystem."
- **Answer in the first two sentences under it.** Then add the detail.
- **Put comparable facts in tables.** Prices, limits, supported countries, sizes and return windows are what an answer quotes.
- **Name sources and dates.** "Updated October 2026" and a link to the standard you cite make a passage safer to quote.
- **Say who it's not for.** An honest limit is a detail answers repeat, and it builds the trust that lets them cite you on the rest.

We go deeper on passage-level writing in [how to get cited by AI](/blog/how-to-get-cited-by-ai/).

## Step 6: Track the family, not the head prompt

If fan-out decides who gets cited, a dashboard that tracks only head prompts measures the wrong thing. AI visibility tools track only the prompts you give them, so a family you never entered is a family you'll never see move.

At Earleads, priority questions are sampled many times per engine and reported as an appearance rate, never as a rank, and the set is re-measured twice a month. Answers vary from run to run, so a single check says little. Apply the same idea to fan-out:

- **Track the head question and its key sub-questions** as separate prompts, grouped by family.
- **Report appearance by family.** "We appear on the pricing and integration branches, not on comparisons" is a plan. "We rank third for the head prompt" isn't.
- **Watch the pages that own each branch.** Google counts AI Overviews and AI Mode traffic in the Search Console Performance report, inside the "Web" search type ([Google Search Central](https://developers.google.com/search/docs/appearance/ai-features)). Look at impressions and queries for each spoke page, not just the hub.
- **Check how you're described.** Being named as "a basic option for small teams" on a comparison branch is a different problem from being absent.

For the formula behind appearance rates, see [AI share of voice](/blog/ai-share-of-voice/). For tying citations to pipeline, see [AI search analytics](/blog/ai-search-analytics/).

## Query fan-out for B2B vs B2C

The mechanism is the same. What fans out is different.

**B2B questions are long and layered.** A buyer asks about a category, a company size, a country and a tool they already use, all at once. That fans out deep into integration docs, security pages, pricing by seat and named comparisons. The branches that decide shortlists are often dull ones, like data residency or a specific integration, and they're the ones nobody wrote a page for. Our guide to [GEO for B2B](/blog/geo-for-b2b/) covers how AI shortlists vendors.

**B2C questions lean on fit and experience.** "Best running shoes for flat feet on trails" fans out into fit, durability, sizing against another brand and reviews from people with the same problem. Product pages and size guides own the facts. Reviews and community threads own whether people liked it. For a shop, structured product data and accurate listings carry the fact branches. Our guide on [AI shopping agents](/blog/is-your-store-visible-to-ai-shopping-agents/) covers that side.

In both cases, the first-person branches ("did it work for you") are won off your site, and the factual branches are won on it.

## Common query fan-out mistakes

**Treating a fan-out tool's output as truth.** It's a simulation from a model, not Google's log. Run it more than once, compare it with what buyers ask and keep what recurs.

**Publishing a page for every variant.** Near-duplicates split one answer. Merge variants into the page that already ranks and keep spokes for sub-questions with a different answer.

**Covering every branch on your own site.** "Is it any good" and "has anyone switched" are answered by other people. Writing your own version doesn't make it a third-party opinion.

**Tracking only the head prompt.** If the report shows one prompt per topic, it can't tell you which branch you're losing.

**Burying facts.** Prices in images, specs in PDFs, limits behind a tab that loads with JavaScript. A sub-search can't quote what it can't read.

**Chasing fan-out before basics.** Google is clear that a supporting link must be indexed and eligible for a snippet. If key pages aren't, start there. An [AI visibility audit](/blog/what-an-ai-visibility-audit-should-show-you/) should show you which of these problems you have before anyone starts writing, and you can [request one here](/#audit).

## Frequently asked questions

### What is query fan-out in simple terms?

It's what happens between your question and the AI answer. Google says AI Mode breaks a question into subtopics and issues many searches at the same time, then writes one response from what those searches find. So a single question like which payroll tool fits a small company quietly becomes searches about price, integrations, reviews and alternatives, and each of those has its own winners.

### Does query fan-out happen in AI Overviews or only in AI Mode?

Both. Google's Search Central documentation, last updated December 10, 2025, says AI Overviews and AI Mode may both use a query fan-out technique. Google also says Deep Search in AI Mode uses the same technique at a larger scale and can issue hundreds of searches. The word may matters, because not every query fans out the same way.

### Is there a tool that shows the real fan-out queries?

No tool sees Google's live fan-out for a real user. Query fan-out tools simulate it, usually by asking a Gemini model with search grounding which searches it would run. The Gemini API documents that the model generates one or multiple search queries and returns them. Treat any list as a set of hypotheses, run it more than once and keep the sub-questions that keep coming back.

### Do I need a separate page for every fan-out query?

No. A sub-question earns its own page when it has a different answer and a buyer who could act on it, like pricing, a named comparison or a migration. Spelling variants and near-synonyms belong on one page. Splitting one answer across four thin pages makes each one weaker. Most sub-questions are a well-labelled section on the hub page.

### Is AI Mode SEO different from normal SEO?

Less than people sell it as. Google says there are no additional requirements and no special optimizations needed to appear in AI Overviews or AI Mode, and no AI text files or special schema. What changes is the unit of planning. You plan for the family of questions around a buyer question, not a single keyword, and you check that every branch has a page worth citing.

### How do I measure visibility across fan-out?

Track the family, not just the head prompt. Ask the main question and its key sub-questions in the engines your buyers use, repeat each priority question many times, and report how often you appear. In Search Console, Google counts AI Mode and AI Overviews traffic inside the Web search type of the Performance report, so watch the pages that own each sub-question.